Recursively dir() a python object to find values of a certain type or with a certain value

Question

I have a complex Python data structure (if it matters, it's a large music21 Score object) which will not pickle due to the presence of a weakref somewhere deep inside the object structure. I've debugged such problems with the stack trace and python debugger before, but it's always a big pain. Is there a tool which runs dir() recursively on all attributes of an object, finding objects hidden in lists, tuples, dicts, etc., and returns those that match a certain value (a lambda function or something like that). A big problem is recursive references, so some sort of memo function (like copy.deepcopy uses) is needed. I tried:

import weakref
def findWeakRef(streamObj, memo=None):
    weakRefList = []
    if memo is None:
        memo = {}
    for x in dir(streamObj):
        xValue = getattr(streamObj, x)
        if id(xValue) in memo:
            continue
        else:
            memo[id(xValue)] = True
        if type(xValue) is weakref.ref:
            weakRefList.append(x, xValue, streamObj)
        if hasattr(xValue, "__iter__"):
            for i in xValue:
                if id(i) in memo:
                    pass
                else:
                    memo[id(i)] = True
                    weakRefList.extend(findWeakRef(i), memo)
        else:
            weakRefList.extend(findWeakRef(xValue), memo)
    return weakRefList

I can probably continue plugging holes in this (the iter isn't what I'd want for dicts, for instance), but before I throw more time into it, wondering if someone knows an easier answer. It could be a pretty useful general tool.

Leadership

I haven't seen a ready-made solution yet. Maybe gc.get_referents instead of dir gets you a little farther. – Disaffection

It does cut down on the cruft (eq, etc.) though at a price that the format of what's returned changes by the type of object, so it'd probably be a faster solution, but not simpler. It too doesn't seem to support recursion. Thanks! – Leadership

Have you consider subclassing the pickle.Pickler class? The source is included in .../Lib/pickle.py. Doing so should allow you to reuse a lot of code and trap PickleErrors to do what you describe -- as well as piggy-back off of the well-established pickling-protocol Python already has in place. – Forelli

Good comment. I needed this script for something else as well, but that seems like it could be a great way of doing it. – Leadership

Answer 1

This seems to be the start of an answer. I had to backport some items from the Python 3.2 inspect.getattr_static to make it work so it didn't call properties that just kept generating new objects. Here's the code I came up with:

#-------------------------------------------------------------------------------
# Name:         treeYield.py
# Purpose:      traverse a complex datastructure and yield elements
#               that fit a given criteria
#
# Authors:      Michael Scott Cuthbert
#
# Copyright:    Copyright © 2012 Michael Scott Cuthbert
# License:      CC-BY
#-------------------------------------------------------------------------------
import types

class TreeYielder(object):
    def __init__(self, yieldValue = None):
        '''
        `yieldValue` should be a lambda function that
        returns True/False or a function/method call that
        will be passed the value of a current attribute
        '''        
        self.currentStack = []
        self.yieldValue = yieldValue
        self.stackVals = []
        t = types
        self.nonIterables = [t.IntType, t.StringType, t.UnicodeType, t.LongType,
                             t.FloatType, t.NoneType, t.BooleanType]

    def run(self, obj, memo = None):
        '''
        traverse all attributes of an object looking
        for subObjects that meet a certain criteria.
        yield them.

        `memo` is a dictionary to keep track of objects
        that have already been seen

        The original object is added to the memo and
        also checked for yieldValue
        '''
        if memo is None:
            memo = {}
        self.memo = memo
        if id(obj) in self.memo:
            self.memo[id(obj)] += 1
            return
        else:
            self.memo[id(obj)] = 1

        if self.yieldValue(obj) is True:
            yield obj


        ### now check for sub values...
        self.currentStack.append(obj)

        tObj = type(obj)
        if tObj in self.nonIterables:
            pass
        elif tObj == types.DictType:
            for keyX in obj:
                dictTuple = ('dict', keyX)
                self.stackVals.append(dictTuple)
                x = obj[keyX]
                for z in self.run(x, memo=memo):
                    yield z
                self.stackVals.pop()

        elif tObj in [types.ListType, types.TupleType]:
            for i,x in enumerate(obj):
                listTuple = ('listLike', i)
                self.stackVals.append(listTuple)
                for z in self.run(x, memo=memo):
                    yield z
                self.stackVals.pop()

        else: # objects or uncaught types...
            ### from http://bugs.python.org/file18699/static.py
            try:
                instance_dict = object.__getattribute__(obj, "__dict__")
            except AttributeError:
                ## probably uncaught static object
                return

            for x in instance_dict:
                try:
                    gotValue = object.__getattribute__(obj, x)
                except: # ?? property that relies on something else being set.
                    continue
                objTuple = ('getattr', x)
                self.stackVals.append(objTuple)
                try:
                    for z in self.run(gotValue, memo=memo):
                        yield z
                except RuntimeError:
                    raise Exception("Maximum recursion on:\n%s" % self.currentLevel())
                self.stackVals.pop()                

        self.currentStack.pop()

    def currentLevel(self):
        currentStr = ""
        for stackType, stackValue in self.stackVals:
            if stackType == 'dict':
                if isinstance(stackValue, str):
                    currentStr += "['" + stackValue + "']"
                elif isinstance(stackValue, unicode):
                    currentStr += "[u'" + stackValue + "']"
                else: # numeric key...
                    currentStr += "[" + str(stackValue) + "]"
            elif stackType == 'listLike':
                currentStr += "[" + str(stackValue) + "]"
            elif stackType == 'getattr':
                currentStr += ".__getattribute__('" + stackValue + "')"
            else:
                raise Exception("Cannot get attribute of type %s" % stackType)
        return currentStr

This code lets you run something like this:

class Mock(object):
    def __init__(self, mockThing, embedMock = True):
        self.abby = 30
        self.mocker = mockThing
        self.mockList = [mockThing, mockThing, 40]
        self.embeddedMock = None
        if embedMock is True:
            self.embeddedMock = Mock(mockThing, embedMock = False)

mockType = lambda x: x.__class__.__name__ == 'Mock'

subList = [100, 60, -2]
myList = [5, 20, [5, 12, 17], 30, {'hello': 10, 'goodbye': 22, 'mock': Mock(subList)}, -20, Mock(subList)]
myList.append(myList)

ty = TreeYielder(mockType)
for val in ty.run(myList):
    print(val, ty.currentLevel())

And get:

(<__main__.Mock object at 0x01DEBD10>, "[4]['mock']")
(<__main__.Mock object at 0x01DEF370>, "[4]['mock'].__getattribute__('embeddedMock')")
(<__main__.Mock object at 0x01DEF390>, '[6]')
(<__main__.Mock object at 0x01DEF3B0>, "[6].__getattribute__('embeddedMock')")

Or run:

high = lambda x: isinstance(x, (int, float)) and x > 10
ty = TreeYielder(high)
for val in ty.run(myList):
    print(val, ty.currentLevel())

And get:

(20, '[1]')
(12, '[2][1]')
(17, '[2][2]')
(30, '[3]')
(22, "[4]['goodbye']")
(100, "[4]['mock'].__getattribute__('embeddedMock').__getattribute__('mocker')[0]")
(60, "[4]['mock'].__getattribute__('embeddedMock').__getattribute__('mocker')[1]")
(40, "[4]['mock'].__getattribute__('embeddedMock').__getattribute__('mockList')[2]")

I'm still trying to figure out why .abby isn't found, but I figure it's worth posting even at this point, since it's much more on the right track than I was when I started.

Leadership

Answer 2

Here is a simpler solution that is somewhat naive. I.e. just a depth first search under the attribute tree. If its a primitive then stop , otherwise go deeper in the tree. It will get you the call tree and the value at the leaf.

def recursive_dir(obj, path):
    if ((obj!=None) and (not isinstance(obj, (str,float,int,list,dict,set)))):
        for attr, val in obj.__dict__.iteritems():
            temp_path = path[:]
            temp_path.append(attr)
            recursive_dir(getattr(obj, attr), temp_path)
    else:
        print (path, "--->", obj)
        print("")
recursive_dir(x,[])

Carefree

Great simplification though looking at the con ents of iterables like lists was also important for me to find those pesky weakrefs. – Leadership